Typical Features of OSI Layer 3
A protocol that defines routing and addressing is considered to be a network layer, or Layer 3, protocol. OSI does define a unique Layer 3 protocol called Connectionless Network Services (CLNS), but, as usual with OSI protocols, you rarely see it in networks today. However, you will see many other protocols that perform the OSI Layer 3 functions of routing and addressing, such as the Internet Protocol (IP), Novell Internetwork Packet Exchange (IPX), or AppleTalk Dynamic Data Routing (DDR).
The network layer protocols have many similarities, regardless of what Layer 3 protocol is used. In this section, network layer (Layer 3) addressing is covered in enough depth to describe IP, IPX, and AppleTalk addresses. Also, now that data link layer and network layer addresses have been covered in this book, this section undertakes a comparison between the two.
Routing (Path Selection)
Routing focuses on the end-to-end logic of forwarding data. Figure 5-1 shows a simple example of how routing works. The logic seen in the figure is relatively simple. For PC1 to send data to PC2, it must send something to R1, when sends it to R2, then on to R3, and finally to PC2. However, the logic used by each device along the path varies slightly.
PC1's Logic: Sending Data to a Nearby Router
In this example, PC1 has some data to send data to PC2. Because PC2 is not on the same Ethernet as PC1, PC1 needs to send the packet to a router that is attached to the same Ethernet as PC1. The sender sends a data-link frame across the medium to the nearby router; this frame includes the packet in the data portion of the frame. That frame uses data link layer (Layer 2) addressing in the data-link header to ensure that the nearby router receives the frame.
Figure 5-1 Routing Logic: PC1 Sending to PC2
The main point here is that the originator of the data does not know much about the networkjust how to get the data to some nearby router. In the post office analogy, it's like knowing how to get to the local post office, but nothing more. Likewise, PC1 needs to know only how to get the packet to R1.
R1 and R2's Logic: Routing Data Across the Network
R1 and R2 both use the same general process to route the packet. The routing table for any particular network layer protocol contains a list of network layer address groupings. Instead of a single entry in the routing table per individual destination address, there is one entry per group. The router compares the destination network layer address in the packet to the entries in the routing table, and a match is made. This matching entry in the routing table tells this router where to forward the packet next. The words in the bubbles in Figure 5-1 point out this basic logic.
The concept of network layer address grouping is similar to the U.S. ZIP code system. Everyone living in the same vicinity is in the same ZIP code, and the postal sorters just look for the ZIP codes, ignoring the rest of the address. Likewise, in Figure 5-1, everyone in this network whose IP address starts with 168.1 is on the Token Ring on which PC2 resides, so the routers can just have one routing table entry that means "all addresses that start with 168.1."
Any intervening routers repeat the same process. The destination network layer (Layer 3) address in the packet identifies the group in which the destination resides. The routing table is searched for a matching entry, which tells this router where to forward the packet next. Eventually, the packet is delivered to the router connected to the network or subnet of the destination host (R3), as previously shown in Figure 5-1.
R3's Logic: Delivering Data to the End Destination
The final router in the path, R3, uses almost the exact same logic as R1 and R2, but with one minor difference. R3 needs to forward the packet directly to PC2, not to some other router. On the surface, that difference seems insignificant. In the next section, when you read about how the network layer uses the data link layer, the significance of the difference will become obvious.
Network Layer Interaction with the Data Link Layer
In Figure 5-1, four different types of data links were used to deliver the data. When the network layer protocol is processing the packet, it decides to send the packet out the appropriate network interface. Before the actual bits can be placed onto that physical interface, the network layer must hand off the packet to the data link layer protocols, which, in turn, ask the physical layer to actually send the data. And as was described in Chapter 3, "Data Link Layer Fundamentals: Ethernet LANs," the data link layer adds the appropriate header and trailer to the packet, creating a frame, before sending the frames over each physical network.
The routing process forwards the packet, and only the packet, from end-to-end through the network, discarding data link headers and trailers along the way. The network layer processes deliver the packet end-to-end, using successive data-link headers and trailers just to get the packet to the next router or host in the path. Each successive data link layer just gets the packet from one device to the next. Figure 5-2 shows the same diagram as Figure 5-1 but includes the concepts behind encapsulation.
Figure 5-2 Network Layer and Data Link Layer Encapsulation
Because the routers build new data-link headers and trailers (trailers not shown in figure),and because the new headers contain data-link addresses, the PCs and routers must have some way to decide what data-link addresses to use. An example of how the router determines which data-link address to use is the IP Address Resolution Protocol (ARP). ARP is used to dynamically learn the data-link address of an IP host connected to a LAN. You will read more about ARP later in this chapter.
In short, the process of routing forwards Layer 3 packets, also called Layer 3 protocol data units (L3 PDUs), based on the destination Layer 3 address in the packet. The process uses the data link layer to encapsulate the Layer 3 packets into Layer 2 frames for transmission across each successive data link.
Network Layer (Layer 3) Addressing
One key feature of network layer addresses is that they were designed to allow logical grouping of addresses. In other words, something about the numeric value of an address implies a group or set of addresses, all of which are considered to be in the same grouping. In TCP/IP, this group is called a network or a subnet. In IPX, it is called a network. In AppleTalk, the grouping is called a cable range. These groupings work just like U.S.P.S. ZIP codes, allowing the routers (mail sorters) to speedily route (sort) lots of packets (letters).
Just like postal street addresses, network layer addresses are grouped based on physical location in a network. The rules differ for some network layer protocols, but the grouping concept is identical for IP, IPX, and AppleTalk. In each of these network layer protocols, all devices on opposite sides of a router must be in a different Layer 3 group, just like in the examples earlier in this chapter.
Routing relies on the fact that Layer 3 addresses are grouped together. The routing tables for each network layer protocol can have one entry for the group, not one entry for each individual address. Imagine an Ethernet with 100 TCP/IP hosts. A router needing to forward packets to any of those hosts needs only one entry in its IP routing table. This basic fact is one of the key reasons that routers can scale to allow tens and hundreds of thousands of devices. It's very similar to the U.S.P.S. ZIP code systemit would be ridiculous to have people in the same ZIP code live somewhere far away from each other, or to have next-door neighbors be in different zip codes. The poor postman would spend all his time driving and flying around the country! Similarly, to make routing more efficient, network layer protocols group addresses together.
With that in mind, most network layer (Layer 3) addressing schemes were created with the following goals:
The address space should be large enough to accommodate the largest network for which the designers imagined the protocol would be used.
The addresses should allow for unique assignment.
The address structure should have some grouping implied so that many addresses are considered to be in the same group.
Dynamic address assignment for clients is desired.
The U.S. Postal Service analogy also works well as a comparison to how IP network numbers are assigned. Instead of getting involved with every small community's plans for what to name new streets, the post service simply has a nearby office with a ZIP code. If that local town wants to add streets, the rest of the post offices in the country already are prepared because they just forward letters based on the ZIP code, which they already know. The only postal employees who care about the new streets are the people in the local post office. It is the local postmaster's job to assign a mail carrier to deliver and pick up mail on any new streets.
Also, you can have duplicate local street addresses, as long as they are in different ZIP codes, and it all still works. There might be hundreds of Main streets in different ZIP codes, but as long as there is just one per ZIP code, the address is unique. Layer 3 network addresses follow the same conceptas long as the entire Layer 3 address is unique compared to the other Layer 3 addresses, all is well.
Example Layer 3 Address Structures
Each Layer 3 address structure contains at least two parts. One (or more) part at the beginning of the address works like the ZIP code and essentially identifies the grouping. All instances of addresses with the same value in these first bits of the address are considered to be in the same groupfor example, the same IP subnet or IPX network or AppleTalk cable range. The last part of the address acts as a local address, uniquely identifying that device in that particular group. Table 5-2 outlines several Layer 3 address structures.
Table 5-2 Layer 3 Address Structures
Size of Address in Bits
Name and Size of Grouping Field in Bits
Name and Size of Local Address Field in Bits
Network or subnet (variable, between 8 and 30 bits)
Host (variable, between 2 and 24 bits)
Many formats, many sizes
Domain-specific part (DSPtypically 56, including NSAP)
*Consecutively numbered values in this field can be combined into one group, called a cable range.
Conveniently, the routing tables in the example based on Figure 5-2 had the correct routing information already in their routing tables. In most cases, these entries are built dynamically by use of a routing protocol. Routing protocols learn about all the locations of the network layer "groups" in a network and advertise the locations of the groups. As a result, each router can build a good routing table dynamically. Routing protocols define message formats and procedures, just like any other protocol. The end goal of each routing protocol is to fill the routing table with all known destination groups and with the best route to reach each group.
The terminology relating to routing protocols sometimes can get in the way. A routing protocol learns routes and puts those routes in a routing table. A routed protocol is the type of packet forwarded, or routed, through a network. In Figures 5-1 and 5-2, the figures represent how IP packets are routed, so IP would be the routed protocol. If the routers used the Routing Information Protocol (RIP) to learn the routes, then RIP would be the routing protocol.
Later in this chapter, the section titled "IP Routing Protocols" shows a detailed example of how routing protocols learn routes.